xref: /openbmc/linux/arch/x86/kernel/cpu/mshyperv.c (revision 670c04ad)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * HyperV  Detection code.
4  *
5  * Copyright (C) 2010, Novell, Inc.
6  * Author : K. Y. Srinivasan <ksrinivasan@novell.com>
7  */
8 
9 #include <linux/types.h>
10 #include <linux/time.h>
11 #include <linux/clocksource.h>
12 #include <linux/init.h>
13 #include <linux/export.h>
14 #include <linux/hardirq.h>
15 #include <linux/efi.h>
16 #include <linux/interrupt.h>
17 #include <linux/irq.h>
18 #include <linux/kexec.h>
19 #include <linux/i8253.h>
20 #include <linux/random.h>
21 #include <asm/processor.h>
22 #include <asm/hypervisor.h>
23 #include <asm/hyperv-tlfs.h>
24 #include <asm/mshyperv.h>
25 #include <asm/desc.h>
26 #include <asm/idtentry.h>
27 #include <asm/irq_regs.h>
28 #include <asm/i8259.h>
29 #include <asm/apic.h>
30 #include <asm/timer.h>
31 #include <asm/reboot.h>
32 #include <asm/nmi.h>
33 #include <clocksource/hyperv_timer.h>
34 #include <asm/numa.h>
35 
36 /* Is Linux running as the root partition? */
37 bool hv_root_partition;
38 /* Is Linux running on nested Microsoft Hypervisor */
39 bool hv_nested;
40 struct ms_hyperv_info ms_hyperv;
41 
42 #if IS_ENABLED(CONFIG_HYPERV)
43 static inline unsigned int hv_get_nested_reg(unsigned int reg)
44 {
45 	if (hv_is_sint_reg(reg))
46 		return reg - HV_REGISTER_SINT0 + HV_REGISTER_NESTED_SINT0;
47 
48 	switch (reg) {
49 	case HV_REGISTER_SIMP:
50 		return HV_REGISTER_NESTED_SIMP;
51 	case HV_REGISTER_SIEFP:
52 		return HV_REGISTER_NESTED_SIEFP;
53 	case HV_REGISTER_SVERSION:
54 		return HV_REGISTER_NESTED_SVERSION;
55 	case HV_REGISTER_SCONTROL:
56 		return HV_REGISTER_NESTED_SCONTROL;
57 	case HV_REGISTER_EOM:
58 		return HV_REGISTER_NESTED_EOM;
59 	default:
60 		return reg;
61 	}
62 }
63 
64 u64 hv_get_non_nested_register(unsigned int reg)
65 {
66 	u64 value;
67 
68 	if (hv_is_synic_reg(reg) && hv_isolation_type_snp())
69 		hv_ghcb_msr_read(reg, &value);
70 	else
71 		rdmsrl(reg, value);
72 	return value;
73 }
74 EXPORT_SYMBOL_GPL(hv_get_non_nested_register);
75 
76 void hv_set_non_nested_register(unsigned int reg, u64 value)
77 {
78 	if (hv_is_synic_reg(reg) && hv_isolation_type_snp()) {
79 		hv_ghcb_msr_write(reg, value);
80 
81 		/* Write proxy bit via wrmsl instruction */
82 		if (hv_is_sint_reg(reg))
83 			wrmsrl(reg, value | 1 << 20);
84 	} else {
85 		wrmsrl(reg, value);
86 	}
87 }
88 EXPORT_SYMBOL_GPL(hv_set_non_nested_register);
89 
90 u64 hv_get_register(unsigned int reg)
91 {
92 	if (hv_nested)
93 		reg = hv_get_nested_reg(reg);
94 
95 	return hv_get_non_nested_register(reg);
96 }
97 EXPORT_SYMBOL_GPL(hv_get_register);
98 
99 void hv_set_register(unsigned int reg, u64 value)
100 {
101 	if (hv_nested)
102 		reg = hv_get_nested_reg(reg);
103 
104 	hv_set_non_nested_register(reg, value);
105 }
106 EXPORT_SYMBOL_GPL(hv_set_register);
107 
108 static void (*vmbus_handler)(void);
109 static void (*hv_stimer0_handler)(void);
110 static void (*hv_kexec_handler)(void);
111 static void (*hv_crash_handler)(struct pt_regs *regs);
112 
113 DEFINE_IDTENTRY_SYSVEC(sysvec_hyperv_callback)
114 {
115 	struct pt_regs *old_regs = set_irq_regs(regs);
116 
117 	inc_irq_stat(irq_hv_callback_count);
118 	if (vmbus_handler)
119 		vmbus_handler();
120 
121 	if (ms_hyperv.hints & HV_DEPRECATING_AEOI_RECOMMENDED)
122 		apic_eoi();
123 
124 	set_irq_regs(old_regs);
125 }
126 
127 void hv_setup_vmbus_handler(void (*handler)(void))
128 {
129 	vmbus_handler = handler;
130 }
131 
132 void hv_remove_vmbus_handler(void)
133 {
134 	/* We have no way to deallocate the interrupt gate */
135 	vmbus_handler = NULL;
136 }
137 
138 /*
139  * Routines to do per-architecture handling of stimer0
140  * interrupts when in Direct Mode
141  */
142 DEFINE_IDTENTRY_SYSVEC(sysvec_hyperv_stimer0)
143 {
144 	struct pt_regs *old_regs = set_irq_regs(regs);
145 
146 	inc_irq_stat(hyperv_stimer0_count);
147 	if (hv_stimer0_handler)
148 		hv_stimer0_handler();
149 	add_interrupt_randomness(HYPERV_STIMER0_VECTOR);
150 	apic_eoi();
151 
152 	set_irq_regs(old_regs);
153 }
154 
155 /* For x86/x64, override weak placeholders in hyperv_timer.c */
156 void hv_setup_stimer0_handler(void (*handler)(void))
157 {
158 	hv_stimer0_handler = handler;
159 }
160 
161 void hv_remove_stimer0_handler(void)
162 {
163 	/* We have no way to deallocate the interrupt gate */
164 	hv_stimer0_handler = NULL;
165 }
166 
167 void hv_setup_kexec_handler(void (*handler)(void))
168 {
169 	hv_kexec_handler = handler;
170 }
171 
172 void hv_remove_kexec_handler(void)
173 {
174 	hv_kexec_handler = NULL;
175 }
176 
177 void hv_setup_crash_handler(void (*handler)(struct pt_regs *regs))
178 {
179 	hv_crash_handler = handler;
180 }
181 
182 void hv_remove_crash_handler(void)
183 {
184 	hv_crash_handler = NULL;
185 }
186 
187 #ifdef CONFIG_KEXEC_CORE
188 static void hv_machine_shutdown(void)
189 {
190 	if (kexec_in_progress && hv_kexec_handler)
191 		hv_kexec_handler();
192 
193 	/*
194 	 * Call hv_cpu_die() on all the CPUs, otherwise later the hypervisor
195 	 * corrupts the old VP Assist Pages and can crash the kexec kernel.
196 	 */
197 	if (kexec_in_progress && hyperv_init_cpuhp > 0)
198 		cpuhp_remove_state(hyperv_init_cpuhp);
199 
200 	/* The function calls stop_other_cpus(). */
201 	native_machine_shutdown();
202 
203 	/* Disable the hypercall page when there is only 1 active CPU. */
204 	if (kexec_in_progress)
205 		hyperv_cleanup();
206 }
207 
208 static void hv_machine_crash_shutdown(struct pt_regs *regs)
209 {
210 	if (hv_crash_handler)
211 		hv_crash_handler(regs);
212 
213 	/* The function calls crash_smp_send_stop(). */
214 	native_machine_crash_shutdown(regs);
215 
216 	/* Disable the hypercall page when there is only 1 active CPU. */
217 	hyperv_cleanup();
218 }
219 #endif /* CONFIG_KEXEC_CORE */
220 #endif /* CONFIG_HYPERV */
221 
222 static uint32_t  __init ms_hyperv_platform(void)
223 {
224 	u32 eax;
225 	u32 hyp_signature[3];
226 
227 	if (!boot_cpu_has(X86_FEATURE_HYPERVISOR))
228 		return 0;
229 
230 	cpuid(HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS,
231 	      &eax, &hyp_signature[0], &hyp_signature[1], &hyp_signature[2]);
232 
233 	if (eax < HYPERV_CPUID_MIN || eax > HYPERV_CPUID_MAX ||
234 	    memcmp("Microsoft Hv", hyp_signature, 12))
235 		return 0;
236 
237 	/* HYPERCALL and VP_INDEX MSRs are mandatory for all features. */
238 	eax = cpuid_eax(HYPERV_CPUID_FEATURES);
239 	if (!(eax & HV_MSR_HYPERCALL_AVAILABLE)) {
240 		pr_warn("x86/hyperv: HYPERCALL MSR not available.\n");
241 		return 0;
242 	}
243 	if (!(eax & HV_MSR_VP_INDEX_AVAILABLE)) {
244 		pr_warn("x86/hyperv: VP_INDEX MSR not available.\n");
245 		return 0;
246 	}
247 
248 	return HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS;
249 }
250 
251 #ifdef CONFIG_X86_LOCAL_APIC
252 /*
253  * Prior to WS2016 Debug-VM sends NMIs to all CPUs which makes
254  * it difficult to process CHANNELMSG_UNLOAD in case of crash. Handle
255  * unknown NMI on the first CPU which gets it.
256  */
257 static int hv_nmi_unknown(unsigned int val, struct pt_regs *regs)
258 {
259 	static atomic_t nmi_cpu = ATOMIC_INIT(-1);
260 
261 	if (!unknown_nmi_panic)
262 		return NMI_DONE;
263 
264 	if (atomic_cmpxchg(&nmi_cpu, -1, raw_smp_processor_id()) != -1)
265 		return NMI_HANDLED;
266 
267 	return NMI_DONE;
268 }
269 #endif
270 
271 static unsigned long hv_get_tsc_khz(void)
272 {
273 	unsigned long freq;
274 
275 	rdmsrl(HV_X64_MSR_TSC_FREQUENCY, freq);
276 
277 	return freq / 1000;
278 }
279 
280 #if defined(CONFIG_SMP) && IS_ENABLED(CONFIG_HYPERV)
281 static void __init hv_smp_prepare_boot_cpu(void)
282 {
283 	native_smp_prepare_boot_cpu();
284 #if defined(CONFIG_X86_64) && defined(CONFIG_PARAVIRT_SPINLOCKS)
285 	hv_init_spinlocks();
286 #endif
287 }
288 
289 static void __init hv_smp_prepare_cpus(unsigned int max_cpus)
290 {
291 #ifdef CONFIG_X86_64
292 	int i;
293 	int ret;
294 #endif
295 
296 	native_smp_prepare_cpus(max_cpus);
297 
298 #ifdef CONFIG_X86_64
299 	for_each_present_cpu(i) {
300 		if (i == 0)
301 			continue;
302 		ret = hv_call_add_logical_proc(numa_cpu_node(i), i, cpu_physical_id(i));
303 		BUG_ON(ret);
304 	}
305 
306 	for_each_present_cpu(i) {
307 		if (i == 0)
308 			continue;
309 		ret = hv_call_create_vp(numa_cpu_node(i), hv_current_partition_id, i, i);
310 		BUG_ON(ret);
311 	}
312 #endif
313 }
314 #endif
315 
316 static void __init ms_hyperv_init_platform(void)
317 {
318 	int hv_max_functions_eax;
319 	int hv_host_info_eax;
320 	int hv_host_info_ebx;
321 	int hv_host_info_ecx;
322 	int hv_host_info_edx;
323 
324 #ifdef CONFIG_PARAVIRT
325 	pv_info.name = "Hyper-V";
326 #endif
327 
328 	/*
329 	 * Extract the features and hints
330 	 */
331 	ms_hyperv.features = cpuid_eax(HYPERV_CPUID_FEATURES);
332 	ms_hyperv.priv_high = cpuid_ebx(HYPERV_CPUID_FEATURES);
333 	ms_hyperv.misc_features = cpuid_edx(HYPERV_CPUID_FEATURES);
334 	ms_hyperv.hints    = cpuid_eax(HYPERV_CPUID_ENLIGHTMENT_INFO);
335 
336 	hv_max_functions_eax = cpuid_eax(HYPERV_CPUID_VENDOR_AND_MAX_FUNCTIONS);
337 
338 	pr_info("Hyper-V: privilege flags low 0x%x, high 0x%x, hints 0x%x, misc 0x%x\n",
339 		ms_hyperv.features, ms_hyperv.priv_high, ms_hyperv.hints,
340 		ms_hyperv.misc_features);
341 
342 	ms_hyperv.max_vp_index = cpuid_eax(HYPERV_CPUID_IMPLEMENT_LIMITS);
343 	ms_hyperv.max_lp_index = cpuid_ebx(HYPERV_CPUID_IMPLEMENT_LIMITS);
344 
345 	pr_debug("Hyper-V: max %u virtual processors, %u logical processors\n",
346 		 ms_hyperv.max_vp_index, ms_hyperv.max_lp_index);
347 
348 	/*
349 	 * Check CPU management privilege.
350 	 *
351 	 * To mirror what Windows does we should extract CPU management
352 	 * features and use the ReservedIdentityBit to detect if Linux is the
353 	 * root partition. But that requires negotiating CPU management
354 	 * interface (a process to be finalized). For now, use the privilege
355 	 * flag as the indicator for running as root.
356 	 *
357 	 * Hyper-V should never specify running as root and as a Confidential
358 	 * VM. But to protect against a compromised/malicious Hyper-V trying
359 	 * to exploit root behavior to expose Confidential VM memory, ignore
360 	 * the root partition setting if also a Confidential VM.
361 	 */
362 	if ((ms_hyperv.priv_high & HV_CPU_MANAGEMENT) &&
363 	    !(ms_hyperv.priv_high & HV_ISOLATION)) {
364 		hv_root_partition = true;
365 		pr_info("Hyper-V: running as root partition\n");
366 	}
367 
368 	if (ms_hyperv.hints & HV_X64_HYPERV_NESTED) {
369 		hv_nested = true;
370 		pr_info("Hyper-V: running on a nested hypervisor\n");
371 	}
372 
373 	/*
374 	 * Extract host information.
375 	 */
376 	if (hv_max_functions_eax >= HYPERV_CPUID_VERSION) {
377 		hv_host_info_eax = cpuid_eax(HYPERV_CPUID_VERSION);
378 		hv_host_info_ebx = cpuid_ebx(HYPERV_CPUID_VERSION);
379 		hv_host_info_ecx = cpuid_ecx(HYPERV_CPUID_VERSION);
380 		hv_host_info_edx = cpuid_edx(HYPERV_CPUID_VERSION);
381 
382 		pr_info("Hyper-V: Host Build %d.%d.%d.%d-%d-%d\n",
383 			hv_host_info_ebx >> 16, hv_host_info_ebx & 0xFFFF,
384 			hv_host_info_eax, hv_host_info_edx & 0xFFFFFF,
385 			hv_host_info_ecx, hv_host_info_edx >> 24);
386 	}
387 
388 	if (ms_hyperv.features & HV_ACCESS_FREQUENCY_MSRS &&
389 	    ms_hyperv.misc_features & HV_FEATURE_FREQUENCY_MSRS_AVAILABLE) {
390 		x86_platform.calibrate_tsc = hv_get_tsc_khz;
391 		x86_platform.calibrate_cpu = hv_get_tsc_khz;
392 	}
393 
394 	if (ms_hyperv.priv_high & HV_ISOLATION) {
395 		ms_hyperv.isolation_config_a = cpuid_eax(HYPERV_CPUID_ISOLATION_CONFIG);
396 		ms_hyperv.isolation_config_b = cpuid_ebx(HYPERV_CPUID_ISOLATION_CONFIG);
397 
398 		if (ms_hyperv.shared_gpa_boundary_active)
399 			ms_hyperv.shared_gpa_boundary =
400 				BIT_ULL(ms_hyperv.shared_gpa_boundary_bits);
401 
402 		pr_info("Hyper-V: Isolation Config: Group A 0x%x, Group B 0x%x\n",
403 			ms_hyperv.isolation_config_a, ms_hyperv.isolation_config_b);
404 
405 		if (hv_get_isolation_type() == HV_ISOLATION_TYPE_SNP)
406 			static_branch_enable(&isolation_type_snp);
407 	}
408 
409 	if (hv_max_functions_eax >= HYPERV_CPUID_NESTED_FEATURES) {
410 		ms_hyperv.nested_features =
411 			cpuid_eax(HYPERV_CPUID_NESTED_FEATURES);
412 		pr_info("Hyper-V: Nested features: 0x%x\n",
413 			ms_hyperv.nested_features);
414 	}
415 
416 #ifdef CONFIG_X86_LOCAL_APIC
417 	if (ms_hyperv.features & HV_ACCESS_FREQUENCY_MSRS &&
418 	    ms_hyperv.misc_features & HV_FEATURE_FREQUENCY_MSRS_AVAILABLE) {
419 		/*
420 		 * Get the APIC frequency.
421 		 */
422 		u64	hv_lapic_frequency;
423 
424 		rdmsrl(HV_X64_MSR_APIC_FREQUENCY, hv_lapic_frequency);
425 		hv_lapic_frequency = div_u64(hv_lapic_frequency, HZ);
426 		lapic_timer_period = hv_lapic_frequency;
427 		pr_info("Hyper-V: LAPIC Timer Frequency: %#x\n",
428 			lapic_timer_period);
429 	}
430 
431 	register_nmi_handler(NMI_UNKNOWN, hv_nmi_unknown, NMI_FLAG_FIRST,
432 			     "hv_nmi_unknown");
433 #endif
434 
435 #ifdef CONFIG_X86_IO_APIC
436 	no_timer_check = 1;
437 #endif
438 
439 #if IS_ENABLED(CONFIG_HYPERV) && defined(CONFIG_KEXEC_CORE)
440 	machine_ops.shutdown = hv_machine_shutdown;
441 	machine_ops.crash_shutdown = hv_machine_crash_shutdown;
442 #endif
443 	if (ms_hyperv.features & HV_ACCESS_TSC_INVARIANT) {
444 		/*
445 		 * Writing to synthetic MSR 0x40000118 updates/changes the
446 		 * guest visible CPUIDs. Setting bit 0 of this MSR  enables
447 		 * guests to report invariant TSC feature through CPUID
448 		 * instruction, CPUID 0x800000007/EDX, bit 8. See code in
449 		 * early_init_intel() where this bit is examined. The
450 		 * setting of this MSR bit should happen before init_intel()
451 		 * is called.
452 		 */
453 		wrmsrl(HV_X64_MSR_TSC_INVARIANT_CONTROL, HV_EXPOSE_INVARIANT_TSC);
454 		setup_force_cpu_cap(X86_FEATURE_TSC_RELIABLE);
455 	}
456 
457 	/*
458 	 * Generation 2 instances don't support reading the NMI status from
459 	 * 0x61 port.
460 	 */
461 	if (efi_enabled(EFI_BOOT))
462 		x86_platform.get_nmi_reason = hv_get_nmi_reason;
463 
464 	/*
465 	 * Hyper-V VMs have a PIT emulation quirk such that zeroing the
466 	 * counter register during PIT shutdown restarts the PIT. So it
467 	 * continues to interrupt @18.2 HZ. Setting i8253_clear_counter
468 	 * to false tells pit_shutdown() not to zero the counter so that
469 	 * the PIT really is shutdown. Generation 2 VMs don't have a PIT,
470 	 * and setting this value has no effect.
471 	 */
472 	i8253_clear_counter_on_shutdown = false;
473 
474 #if IS_ENABLED(CONFIG_HYPERV)
475 	if ((hv_get_isolation_type() == HV_ISOLATION_TYPE_VBS) ||
476 	    (hv_get_isolation_type() == HV_ISOLATION_TYPE_SNP))
477 		hv_vtom_init();
478 	/*
479 	 * Setup the hook to get control post apic initialization.
480 	 */
481 	x86_platform.apic_post_init = hyperv_init;
482 	hyperv_setup_mmu_ops();
483 	/* Setup the IDT for hypervisor callback */
484 	alloc_intr_gate(HYPERVISOR_CALLBACK_VECTOR, asm_sysvec_hyperv_callback);
485 
486 	/* Setup the IDT for reenlightenment notifications */
487 	if (ms_hyperv.features & HV_ACCESS_REENLIGHTENMENT) {
488 		alloc_intr_gate(HYPERV_REENLIGHTENMENT_VECTOR,
489 				asm_sysvec_hyperv_reenlightenment);
490 	}
491 
492 	/* Setup the IDT for stimer0 */
493 	if (ms_hyperv.misc_features & HV_STIMER_DIRECT_MODE_AVAILABLE) {
494 		alloc_intr_gate(HYPERV_STIMER0_VECTOR,
495 				asm_sysvec_hyperv_stimer0);
496 	}
497 
498 # ifdef CONFIG_SMP
499 	smp_ops.smp_prepare_boot_cpu = hv_smp_prepare_boot_cpu;
500 	if (hv_root_partition)
501 		smp_ops.smp_prepare_cpus = hv_smp_prepare_cpus;
502 # endif
503 
504 	/*
505 	 * Hyper-V doesn't provide irq remapping for IO-APIC. To enable x2apic,
506 	 * set x2apic destination mode to physical mode when x2apic is available
507 	 * and Hyper-V IOMMU driver makes sure cpus assigned with IO-APIC irqs
508 	 * have 8-bit APIC id.
509 	 */
510 # ifdef CONFIG_X86_X2APIC
511 	if (x2apic_supported())
512 		x2apic_phys = 1;
513 # endif
514 
515 	/* Register Hyper-V specific clocksource */
516 	hv_init_clocksource();
517 	hv_vtl_init_platform();
518 #endif
519 	/*
520 	 * TSC should be marked as unstable only after Hyper-V
521 	 * clocksource has been initialized. This ensures that the
522 	 * stability of the sched_clock is not altered.
523 	 */
524 	if (!(ms_hyperv.features & HV_ACCESS_TSC_INVARIANT))
525 		mark_tsc_unstable("running on Hyper-V");
526 
527 	hardlockup_detector_disable();
528 }
529 
530 static bool __init ms_hyperv_x2apic_available(void)
531 {
532 	return x2apic_supported();
533 }
534 
535 /*
536  * If ms_hyperv_msi_ext_dest_id() returns true, hyperv_prepare_irq_remapping()
537  * returns -ENODEV and the Hyper-V IOMMU driver is not used; instead, the
538  * generic support of the 15-bit APIC ID is used: see __irq_msi_compose_msg().
539  *
540  * Note: for a VM on Hyper-V, the I/O-APIC is the only device which
541  * (logically) generates MSIs directly to the system APIC irq domain.
542  * There is no HPET, and PCI MSI/MSI-X interrupts are remapped by the
543  * pci-hyperv host bridge.
544  *
545  * Note: for a Hyper-V root partition, this will always return false.
546  * The hypervisor doesn't expose these HYPERV_CPUID_VIRT_STACK_* cpuids by
547  * default, they are implemented as intercepts by the Windows Hyper-V stack.
548  * Even a nested root partition (L2 root) will not get them because the
549  * nested (L1) hypervisor filters them out.
550  */
551 static bool __init ms_hyperv_msi_ext_dest_id(void)
552 {
553 	u32 eax;
554 
555 	eax = cpuid_eax(HYPERV_CPUID_VIRT_STACK_INTERFACE);
556 	if (eax != HYPERV_VS_INTERFACE_EAX_SIGNATURE)
557 		return false;
558 
559 	eax = cpuid_eax(HYPERV_CPUID_VIRT_STACK_PROPERTIES);
560 	return eax & HYPERV_VS_PROPERTIES_EAX_EXTENDED_IOAPIC_RTE;
561 }
562 
563 const __initconst struct hypervisor_x86 x86_hyper_ms_hyperv = {
564 	.name			= "Microsoft Hyper-V",
565 	.detect			= ms_hyperv_platform,
566 	.type			= X86_HYPER_MS_HYPERV,
567 	.init.x2apic_available	= ms_hyperv_x2apic_available,
568 	.init.msi_ext_dest_id	= ms_hyperv_msi_ext_dest_id,
569 	.init.init_platform	= ms_hyperv_init_platform,
570 };
571